computational prediction of protein-protein...

Computational Prediction of Protein-Protein Interactions Enright A.J., Skrabanek L. and Bader G.D.

Computational Prediction ofProtein-Protein InteractionsAnton J. Enright1, Lucy Skrabanek2 and Gary D. Bader1

1 Computational Biology Center, Memorial Sloan-Kettering Cancer Center,307 East 63rd Street, New York, NY 10021, USA.Email: [email protected], [email protected] Ph: (646) 735 8079, Fax: (646) 735 0021

2 Department of Physiology and Biophysics and Institute for Computational Biomedicine,Weill Medical College of Cornell University, 1300 York Avenue, New York, NY 10021, USA.Email: [email protected] Ph: (212) 327 7361, Fax: (212) 860 3369

Abstract

Recently a number of computational approaches have been developed for the predictionof protein-protein interactions. Complete genome sequencing projects have provided thevast amount of information needed for these analyses. These methods utilize thestructural, genomic and biological context of proteins and genes in complete genomes topredict protein interaction networks and functional linkages between proteins. Given thatexperimental techniques remain expensive, time-consuming and labor-intensive, thesemethods represent an important advance in proteomics. Some of these approaches utilizesequence data alone to predict interactions, while others combine multiple computationaland experimental datasets to accurately build protein interaction maps for completegenomes. These methods represent a complementary approach to current high-throughputprojects whose aim is to delineate protein interaction maps in complete genomes. In thischapter, we will describe a number of computational protocols for protein interactionprediction based on the structural, genomic and biological context of proteins in completegenomes, and detail methods for protein interaction network visualization and analysis.

Keywords: Genome context, gene fusion, phylogenetic profiles, geneneighborhood, protein interaction networks, visualization.


1. Introduction

One of the current goals of proteomics is to map the protein interaction networks of alarge number of model organisms (1). Protein-protein interaction information allows thefunction of a protein to be defined by its position in a complex web of interactingproteins. Access to such information will greatly aid biological research and potentiallymake the discovery of novel drug targets much easier. Previously the detection ofprotein-protein interactions was limited to labor-intensive experimental techniques suchas co-immunoprecipitation or affinity chromatography. High-throughput experimentaltechniques such as yeast two-hybrid and mass spectrometry have now also becomeavailable for large-scale detection of protein interactions. These methods however, maynot be generally applicable to all proteins in all organisms, and may also be prone tosystematic error. Recently, a number of complementary computational approaches havebeen developed for the large-scale prediction of protein-protein interactions based onprotein sequence, structure and evolutionary relationships in complete genomes.

Initially computational prediction of protein-protein interactions was strictly limited toproteins whose three-dimensional structures had been determined. These methodspredicted protein-protein interaction based on the structural context of proteins. Recentadvances in complete genome sequencing have however provided a wealth of genomicinformation. It is now possible to establish the genomic context of a given gene in acomplete genome (2,3). A gene is no longer thought of as a single protein-coding entitybut as part of a coordinated network of interacting proteins. The potential for two proteinsto interact is not only specified by the physical and structural properties of theirstructures, but is also encoded at a genomic level. For example, interacting genes aregenerally co-expressed (4-6) (both temporally and spatially). In other words, the fact thattwo proteins have the physical potential to interact is meaningless unless they are presentin the same part of the cell at the same time. Other examples of genomic context includethe co-localization of genes on chromosomes, the complete fusion of pairs of genes,correlated mutations between interacting protein families, and phylogenetic gene profiles.Even in the absence of structural or sequence information, one can detect theevolutionary fingerprints of pairs of interacting proteins from their genomic context. Anumber of these computational approaches also take advantage of high-throughputexperimental information such as gene-expression data, cellular locality and molecularcomplex information (7,8). These hybrid computational approaches exploit both thegenomic and biological context of genes and proteins in complete genomes in order topredict interactions.

In this chapter, we will describe computational methods and resources availablefor protein-protein interaction prediction that exploit the structural, genomic andbiological contexts of proteins in complete genomes. In addition to algorithms andmethods for interaction prediction, a number of useful databases pertaining to protein-protein interaction will be described. These databases combine a large amount of datafrom both computational and experimental techniques. Finally, a number of tools forprotein interaction network visualization and analysis will be described. Methods are


presented in historical order together with online access information. Where available,detailed computational protocols will also be provided for each method.

1.1 Structural Context Approaches

Computational prediction of protein-protein interactions consists of two main areas (i) themapping of protein-protein interactions i.e., determining whether two proteins are likelyto interact, and (ii) the understanding of the mechanism of protein-protein interactionsand the identification of residues in proteins which are involved in those interactions. Thefirst successful computational analyses of protein-protein interactions, used the structuralcontext of proteins in order analyze known protein interaction interfaces in order todetermine physical rules determining protein-protein interaction specificity. Unlike othercomputational methods that use an evolutionary or genomic context to predict interaction,structural approaches tend to be more limited in terms of scale, as only a small proportionof protein sequences have accurate three-dimensional structures deposited in the ProteinDatabank (PDB) (9). However, structural approaches allow for a much more detailedanalysis of protein interactions than the genome-context based approaches. Structuralapproaches can determine, not only whether two proteins interact, but also the physicalcharacteristics of the interaction, and residues (sites) at the protein interface whichmediate the interaction.

The identification of protein interaction sites is important for functional genomics,analysis of metabolic and signal transduction networks and also rational drug design. Thefirst attempt to describe the characteristics of protein interaction sites was undertaken in1975 by Chothia and Janin. With data from only three complexes, they suggested that theresidues which form the interface are closely packed, tend to be hydrophobic and thatcomplementarity may be an important factor in predicting which proteins can interact(10).

Later studies with larger datasets extended and developed their work to try toidentify other characteristics of the interaction site that are sufficiently different from therest of the protein to be identifiable, and thus be predictive. Further analysis of thehydrophobicity distribution of amino acids can be used to predict interaction sites sinceinteracting regions tend to be the most hydrophobic clusters on the surface of the protein(11-14). This type of analysis yields a 60% success rate at predicting interacting sites. Ingeneral, hydrophobic residues such as Leu, Ile, Val, Phe, Tyr and Met are over-represented at interaction sites, whereas polar residues such as Lys, Asp and Glu (but notArg) are under-represented (15,16). Other parameters which have been analyzed for theirimportance in identifying those residues in a protein which form the interaction siteinclude the accessible surface area and residue composition (17,18). It has also becomeapparent that a distinction must be made between different types of complexes.Interaction sites on stable and transient complexes have different properties (18). Arecent study (19) indicates that the the residue composition can be used to identify sixdifferent types of protein-protein interfaces, from domain-domain interfaces in the sameprotein to inter-protein contact surfaces.


Further studies, (18,20) using a six-parameter analysis (solvation potential,residue interface potential, hydrophobicity, planarity, protrusion and accessible surfacearea), have indicated that none of these parameters individually could be definitively usedas a prediction method. Using a combined score from all six parameters yielded accuratepredictions for 66% of 59 structures (21). All interfaces tend to be planar and be surfaceaccessible, the other parameters differed between complex types. Computationalresources for the prediction and analysis of protein-protein interactions in this way aredescribed below (see Methods 2.1. and Table 1).

Shape complementarity is primarily used in docking studies which focus onfinding the best fit of the two interacting proteins using rigid- and soft-body searches (22-24). Electrostatic complementarity between interfaces (Fig. 1) plays an important role indetermining the best fit of two interacting proteins (23). Interfaces between antibody-antigen complexes and transient heterodimers tend to have the least shapecomplementarity, while homodimers, enzyme-inhibitor complexes and permanentheterodimers are the most complementary (18).

However, other research (25) has indicated that the chemico-physical propertiesof interacting surfaces are difficult to distinguish from those of the whole protein surface.Recently, it has been suggested that instead of using patch analysis, it may be better touse interface contacts (19), i.e., residues whose closest atoms are annotated in PDB asbeing less than 6Å apart. They argue that the analysis of surface patches may missslightly buried residues with long side-chains, while other residues identified as beingpart of a patch may in fact not be important, or may not form contacts at all.

Other methods of predicting protein interaction sites include multiple sequencealignment and analysis of amino acid characteristics of neighboring residues using neuralnetworks. Multiple sequence alignments can help identify specific family structureswhich are conserved within a subfamily but differ between subfamilies. These regions areinterpreted as being interaction sites which may endow specificity of ligand interaction(26-28). Two groups (29,30) have trained neural networks with sequence profiles ofspatial neighbors of a target residue with solvent exposure to predict whether a residuewill be part of an interaction site. Both of their methods gave approximately a 70%accurate prediction rate. The validity of using sequence profiles has been verified byresults which demonstrate that the majority of interacting residues are clustered insequence segments of several contacting residues (31).

Recently, methods have been developed to validate predicted protein-proteininteractions against experimentally determined 3D structures (32,33). Given a knownthree-dimensional structure, they map homologs of the interacting proteins onto thestructure and using empirical potentials, test whether the homologous proteins preservethe interactions from the known structure. However, the number of experimentallydetermined structures for complexes is small, and of the 2,590 interactions predicted bylarge-scale methods, only 59 could be mapped onto their set of interacting complexes. Ofthese, 59% had domains that appeared to be in direct contact, thus increasing the


probability that these predicted protein-protein interactions are biologically correct.Computational methods (34) for the prediction of protein-protein interactions based onthis (and other structural approaches) are described below (see Methods 2.1.).

1.2. Genomic Context Approaches

1.2.1. Co-localization

One of the first methods for predicting protein-protein interactions from the genomiccontext of genes utilizes the idea of co-localization, or gene neighborhood (Fig. 2A).Such methods exploit the notion that genes which physically interact (or are functionallyassociated) will be kept in close physical proximity to each other on the genome (35-37).The most apparent case of this phenomenon involves bacterial and archaeal operons,where genes that work together are generally transcribed on the same polycistronicmRNA. In these cases, proteins involved in the same process or pathway are frequentlyencoded on the same polycistronic messenger.

Operons are rare in eukaryotic species (38,39). However, genes involved in thesame biological process or pathway are frequently situated in close genomic proximity(36). It is hence possible to predict functional or physical interaction between genes thatare repeatedly observed in close proximity (e.g. within 500 bp) across many genomes.This method has been successfully used to identify new members of metabolic pathways(36). Like many of the genome-context approaches, this method becomes more powerfulwith larger numbers of genomes. This approach and a number of online resources thatimplement it will be described in detail below (see Methods 2.2.).

1.2.2. Phylogenetic Profiles

A relatively simple, yet powerful, form of genomic context is the co-occurence of pairs ofgenes across multiple genomes. Two of the main driving forces in genome evolution aregene genesis and gene loss (40,41). The fact that a pair of genes remains together acrossmany disparate species represents a concerted evolutionary effort that suggests that thesegenes are functionally associated (i.e. same biological process or pathway) or physicallyinteracting. This criterion is less stringent than that of gene co-localization, where genepairs must not only be present, but also situated close to each other on the genome.Homologous genes can be termed either orthologs or paralogs. In general the termortholog is used to describe genes that are related by a speciation event, i.e. performanalogous functions in different organisms and are related to a single common ancestorgene in an ancestor species. The term paralog is used to describe homologous genes thathave arisen following a gene duplication event, i.e., perform similar functions in the sameorganism. Classifying homologous genes as either paralogs or orthologs is difficult in theabsence of accurate phylogenetic or speciation information (42). Classification of genesin this way allows the inference of a phylogenetic context for a given gene.

The analysis of phylogenetic context in this fashion has been termed phylogeneticprofiling (43). These profiles can be as simple as a binary representation of the presence


or absence of a gene across multiple genomes (43-45) (Fig. 2B). A library of theseprofiles may then be scanned to find genes that exhibit identical (or highly similar)phylogenetic patterns to each other. Pairs of genes detected in this fashion are hencecandidates for physical interaction or functional association. This method has been usednot only to infer physical interaction (43), but also to predict the cellular localization ofgene products (46).

This system is not however without flaw. Firstly, the strength of any inferencemade using such profiles is heavily dependent on the number and distribution of genomesused to build the profile. A pair of genes with similar profiles across many of bacterial,archaeal and eukaryotic genomes are much more likely to interact than genes found to co-occur in a small number of closely related species. Secondly, evolutionary processes suchas lineage-specific gene loss, horizontal gene transfer, non-orthologous gene-displacement (47) and the extensive expansion of many eukaryotic gene families canmake orthology assignment across genomes very difficult. However, given the increasingnumber of completely sequenced genomes, the accuracy of these predictions is expectedto improve over time. The details of this approach and online-resources for phylogeneticprofile-based prediction of protein interaction are described below (see Methods 2.3.).

1.2.3. Gene Fusion

Genome context approaches to the prediction of protein-protein interaction also includethe analysis of gene fusion across complete genomes. This method is complementary toboth co-localization of genes and phylogenetic profiles and uses both gene location andphylogenetic analysis to infer function or interaction. A gene fusion event represents thephysical fusion of two separate parent genes into a single multi-functional gene. This isthe ultimate form of gene co-localization, i.e., interacting genes are not just kept in closeproximity on the genome, but are physically joined into a single entity (Fig. 2A). It hasbeen suggested that the driving force behind these events is to lower the regulational loadof multiple interacting gene products (48). Gene fusion events hence provide an elegantway to computationally detect functional and physical interactions between proteins(48,49).

Gene fusion events are detected by cross-species sequence comparison. Fused(composite) proteins in a given reference genome are detected by searching for un-fusedcomponent protein sequences, that are homologous to the reference protein, but not toeach other. These un-fused query sequences align to different regions of the referenceprotein, indicating that it is a composite protein resulting from a gene fusion event (48).Once again, predictions of this type are complicated by a number of issues. The largesthindrance is the presence of so called promiscuous domains. These domains (such ashelix-turn-helix (HTH) and DnaJ) are highly abundant in eukaryotic organisms. Thedomain complexity of eukaryotic proteins coupled with the presence of promiscuousdomains and large degrees of paralogy can hamper the accurate detection of gene fusionevents (50).


Although the method is not generally applicable to all genes, i.e., it requires thatan observable fusion event can be detected between gene pairs, it has been successfullyapplied to a large number of genomes (including eukaryotes) (51). The basic gene fusiondetection method and online resources such as the AllFUSE database (51), will bedescribed in detail below (see Methods 2.4.).

1.2.4. In-silico two-hybrid

The in-silico two-hybrid (i2h) approach has much in common with the other genome-context approaches, but also indirectly assesses structural properties of proteins thatpotentially interact. It has previously been shown that a mutation in the sequence of oneprotein in a pair of interacting proteins is frequently mirrored by a compensatorymutation in its interacting partner (Fig. 2C). The detection of such correlated mutationscan not only be used to predict protein-protein interactions, but also has the potential toidentify specific residues involved at the interaction sites (52).

Previous analyses (53) involved searching for correlation of residue mutationsbetween sequences in the same protein family alignment (intra-family). The in-silico two-hybrid method extends this approach by searching for such mutations across differentprotein families (interfamily). Prediction of protein-protein interactions using thisapproach is achieved by taking pairs of protein family alignments and concatenatingthese alignments into a single cross-family alignment. A position-specific matrix is thenbuilt from this alignment, and a correlation function is then applied to detect residueswhich are correlated both within and across families. Correlated sites that potentiallyindicate protein interaction are returned with a score. The method suffers due to thecomputational complexity of constructing the large numbers of alignments needed, andpoor quality alignments can dramatically increase noise in the procedure (52). Howeverthe method is similar to the gene fusion approach, as a single accurate prediction betweentwo proteins can infer interaction between all members of both families used. Thisapproach is not discussed in the Methods section, as currently the method is not freelyavailable (see Note 1).

1.3. Biological Context Approaches

High throughput experimental techniques now provide access to a more detailed view ofbiological processes at a genomic level. Gene expression analysis allows one to not onlydetermine which genes are active in a given state, but also sets of genes which are co-regulated in many different states. It has been shown that many interacting proteins areco-expressed according to microarray analyses (4-6). Current gene-expression methodsnow allow for every coding gene of a genome to be placed on a single microarray,allowing the activity of every gene to be monitored across different states or time-points.Although these methods cannot directly be used to determine whether or not two proteinsinteract or not, a number of computational approaches have been developed that use thisinformation towards the prediction of protein-protein interaction and gene regulatorynetworks (4-6). Other high-throughput experimental techniques such as yeast two-hybridspecifically test a bait protein for interactions against a set of prey proteins. The bait and


prey consist of fusion constructs that activate a reporter gene if they interact with eachother. While this method is not as accurate as other techniques such as co-immunoprecipitation, affinity chromatography or gel-overlay assays, it can be appliedrapidly to genome-scale studies of protein-protein interactions.

Many of these high-throughput methods for investigating the biological context ofgenes and proteins are inherently noisy. For example, some proteins in yeast two-hybridassays appear to detect a large number of spurious interactions (false-positives). Geneexpression techniques suffer from a number of problems also, such as cross hybridizationand poor signal-to-noise ratios. Recently however, research has shown that multipledatasets pertaining to the biological context of genes and proteins can be combined usingmachine learning techniques (8). Using Bayesian network analysis it is hence possible tocomputationally combine multiple noisy datasets in such a way that protein-proteininteractions can be more reliably predicted. In this method each source of interactionevidence is compared against samples of known positive (proteins in the same complex)and negative (proteins in different cellular locations) interactions, allowing a statisticalreliability index to be built for each data source. When this information is appliedgenome-wide, a prediction can be made for every protein pair in a genome by combiningdifferent sets of independent evidence according to their calculated reliability. Proteininteractions predicted in this way have been shown to be as reliable as pure experimentaltechniques, while simultaneously covering a larger proportion of genes than mostexperimental methods (8).

A number of available resources for protein-protein interaction data, geneexpression data and Bayesian network analysis of multiple interaction datasets will bedescribed below (see Methods 2.5.).

1.4. Data-sources and Visualization techniques

Computational biology is a data-rich research field. The advent of complete genomesequencing and high-throughput experimental techniques has created an enormousamount of data. In order for these data to be both informative and useful, they must bestored in a sensible and accessible way, and tools must be made available to visualize andexchange this information. A number of initiatives are tackling these problems bycreating freely accessible databases storing a wide variety of biological informationincluding protein-protein interactions. Recently, a number of research groups havecreated visualization tools for biological networks. These tools provide a new way toanalyze protein-protein interaction networks, provide a multitude of different ways torepresent interactions and can overlay other biological information onto these networks.A number of databases that store protein-protein interactions, molecular complexes andpathways will be described later (see Methods 2.6. and Table 1). Finally, we will detailmethods for the visualization and analysis of protein-protein interaction networks. (seeMethods 2.7.).


2. Methods

In this section we will describe computational resources and methods for the predictionof protein-protein interactions. These methods will be detailed in chronological order.Within each section a number of on-line computational resources are described that allowone to perform this type of analysis interactively. Where possible (mostly for genomiccontext based approaches), detailed computational protocols will also be provided.Resources mentioned in this section are further summarized below (see Table 1).

2.1. Structure based prediction of interactions

The Protein-Protein Interaction Server at University College London (UCL) provides asimple web-based interface for exploring protein-protein interaction interfaces, giventhree-dimensional structures (18). This server takes into account the followinginformation for interaction analysis: accessible surface area, planarity, length & breadth,secondary structure, hydrogen bonds, salt bridges, gap volume, gap volume index,bridging water molecules and interface residues. This resource (Table 1) is very usefulfor exploring the protein-protein interaction potential of two protein structures identifiedthrough docking or shape-complementarity.

The structural bioinformatics group at EMBL Heidelberg provides theInterPreTS server for protein-protein interaction prediction (33). Using this resource(Table 1) one can submit pairs of sequences that are then compared to the three-dimensional structures of known protein-protein interactions. This resource utilizes a pre-built Database of Interacting Domains (DBID) and an empirical scoring system to testwhether a sequence pair fits a known three-dimensional structure of an interacting pair ofproteins.

2.2. Gene-neighborhood based interaction prediction

Co-localization of genes across multiple genomes provides a fingerprint that they mayphysically interact (36). Analysis of conserved gene locations across multiple genomes(Fig. 2A) can hence be used to predict protein interaction networks and metabolicpathways (54). A number of excellent resources exist that allow one to determine whethertwo proteins may interact using this approach. The most notable of these are STRING(Search Tool for Recurring Instances of the Neighborhood of Genes) (55) and WIT(What is There?) (56). The STRING database (Fig. 3) provides a web interface givingcomprehensive access to gene neighborhood information (57) for 356,775 genes in 110complete genomes (Fig. 3). Similarly, the Predictome database at Boston University (58)provides a comprehensive web interface to predictions of this type. The WIT databaseprovides access to protein family information, metabolic pathway reconstruction andgene co-localization information. Using these resources allows detailed pre-computedgene neighborhood information to be analyzed for evidence of protein-protein interaction


(Table 1). The actual protocols used for these analyses can vary considerably, a generalprotocol adapted from WIT (56) is described below:

1. In order to assess whether pairs of orthologous genes share a common geneneighborhood across multiple genomes one needs a) protein sequences / genomiclocations and b) orthology mappings between proteins from multiple genomes.

2. Orthology mappings are generated by searching for pairs of close bi-directionalbest hits (PCBBH). These are a specific form of bi-directional best hit (seeMethods 2.3.), a commonly used method for orthology assignment. For a givenpair of proteins a and b in genome X, a bi-directional best hit to genes a’ and b’in genome Y is defined as follows:

a. The best BLAST hit for protein a in genome X is protein a’ in genome Y.b. The best BLAST hit for protein b in genome X is protein b’ in genome Y.c. The genes of proteins a and b are situated within 300bp in genome X.d. The genes of proteins a’ and b’ are situated within 300bp in genome Y.

3. Genes that satisfy the above criteria can be considered as having a conserved geneneighborhood across two genomes. When this procedure is repeated acrossmultiple genomes it becomes possible to identify genes which are significantlyco-localized across many genomes, and are hence likely to either physicallyinteract or be functionally associated.

4. The PCBBH criteria are quite strict, and it is also possible to perform theprocedure using Pairs of Close Homologs (PCHs) (see Note 2).

5. Sets of PCBBHs or PCHs in multiple genomes are typically scored forsignificance based on the number and phylogenetic distribution of genomes inwhich they are co-localized. Phylogenetic distance can be estimated by examininga 16S rRNA phylogenetic tree.

6. A common score (coupling score) for the likelihood that two genes interact basedon summing individual scores from multiple genomes is then calculated.

7. Finally, candidate genes that have significant coupling scores are candidates foreither physical interaction, or functional association.

2.3. Phylogenetic profile based prediction of interaction

Phylogentic profile based prediction of protein interactions (Fig. 2B) has been shown tobe an accurate and widely applicable method. Perhaps the easiest way to utilize thisinformation for prediction of protein interaction is to use precomputed phylogeneticprofiles for proteins of interest. The Clusters of Orthologous Groups (COGs) resource at


the National Center for Biotechnology Information (NCBI) contains large numbers ofprofiles for a variety of bacterial and archaeal organisms and also S. cerevisiae (59,60).Other excellent resources for combined computational predictions of protein interactionsusing phylogenetic profiles are available from the STRING (55) resource at EMBLHeidelberg and from Predictome (58) (Table 1). Using the web interfaces to theseresources, it is relatively straightforward to find groups of proteins with similar oridentical phylogenetic profiles, indicating proteins that physically interact or arefunctionally associated (Fig. 3). For a more detailed analysis of specific proteins ofinterest a general protocol is described below:

1. For each genome to be analyzed, a FASTA sequence file containing all proteinsequences is assembled.

2. All protein sequences in each genome are compared against all other sequencesusing a sequence similarity search algorithm such as BLASTp (61). A variety ofother sequence similarity seach tools could also be used at this step (see Note 3).

3. Orthology between proteins in different genomes is assigned as follows:

• Two proteins (from different genomes) are orthologous if they were eachother’s highest scoring BLAST hit when searched against the other genome.This is frequently referred to as a bidirectional best hit (BBH).

• This process is repeated to assign (if possible) an ortholog for each protein ina given genome, to a protein in all other genomes.

4. All orthology assignments made in this way are stored for post-processing.

5. A phylogenetic profile for a protein can then be constructed by representing thepresence or absence of an ortholog for that protein across all genomes analyzed.Frequently, this is represented by a simple binary vector with ‘1’ indicatingpresence and ‘0’ representing absence of a gene in each genome (Fig. 2B) (seeNote 4).

6. All profiles are compared to all other profiles using a clustering procedure. Adistance measure (such as Pearson correlation of Euclidean distance) betweeneach profile and all other profiles is used to group profiles according to howsimilar they are (see Note 5).

Finally, protein profiles that are highly similar or identical to each other representcandidate proteins that physically or functionally interact.

2.4. Gene fusion prediction of protein interactions


Gene fusion is a relatively common evolutionary phenomenon (51). A detected genefusion between two genes indicates that their protein products may physically interact orbe involved in the same biological process or pathway (48,49). One extreme example ofthis is the aromatic amino acid biosynthesis pathway in S. cerevisiae. In yeast a singlefused gene encodes the entire pathway of these five normally separate genes (48).Prediction of protein interactions using gene fusion has been successful in a number ofareas, including the prediction of novel protein interactions involved in importantbiological processes in Drosophila melanogaster (62).

A comprehensive set of fused genes and inferred protein-protein interactions isavailable from the AllFUSE database (51) at the European Bioinformatics Institute(EBI), the STRING database at EMBL Heidelberg (55) (Fig. 3) and the Predictomedatabase at Boston University (58) (Table 1). Using the AllFUSE resource one cansearch for potential interactions for a given protein sequence from a database of 24complete genomes. A general protocol for gene fusion based prediction of protein-proteininteractions can be described as follows (48):

1. This analysis requires two genomes, a query and a reference. One searches forgene fusion (composite) proteins in the reference genome using protein sequencesfrom the query genome. Sequences from both genomes need to be assembled intoFASTA format for this analysis.

2. Each protein in the query genome is then interactively searched against eachprotein from the reference genome using a sequence similarity search tool such asBLASTp (61) using an expectation-value (E-value) threshold to eliminatesimilarities which may have arisen by chance.

3. All significant similarities detected in this way are then stored in a binary matrixwhich for each protein pair stores ‘1’ for significant similarity or ‘0’ for nodetectable similarity. The matrix may be symmetrified by post-processing with amore sensitive sequence search tool such as Smith-Waterman (63) to clear upambiguities.

4. Finding evidence of a gene fusion event in the reference species extends of theprevious symmetrification problem to one of transitivity (48). In this case onesearches for instances where query proteins A and B match a reference protein C,but do not match each other (i.e. A¤C; B¤C but A≠B). These triangularinequalities are resolved once again by using the more accurate Smith-Watermanalgorithm to double check that no detectable significant similarity exists betweenA and B. Further analysis using alignment geometry can then verify that proteinsA and B are orthologous to different regions of a composite fusion protein but notto each other (64).

5. Candidate fusion proteins detected in this way provide evidence that proteins Aand B may physically interact.


Although this method is not generally applicable to all genes, and suffers from the highlevels of paralogy usually present in eukaryotic genomes (65). This approach has beenshown to have an accuracy as high as 90% and readily detects well known interactingproteins (e.g. tryptophan synthase a and b subunits) and many proteins previously shownto form complexes. As such this method represents a useful way to build interactionnetworks for proteins of interest within and across genomes.

2.5. Prediction of protein interactions from high-throughput biological datasets

Gene expression analysis allows for all genes from a given genome to be placed on asingle microarray, allowing many gene-expression experiments to be carried out rapidlyand in parallel. Recently, efforts have been made to standardize data formats for reportingthe results of gene expression experiments. The Minimum Information About aMicroarray Experiment (MIAME) (66) standard allows different laboratories toeffectively and accurately exchange microarray expression information. Using suchstandards, it has become easier for a number of publicly accessible resources to distributemicroarray data (Table 1).

The Stanford Microarray Database (SMD) (67) provides access to raw data frompublic microarray experiments, as well as a number of software tools for utilizing thisdata. Currently, 140 experiments are indexed in the SMD web resource. The MicroArraygroup at the European Bioinformatics Institute provides ArrayExpress (68), a publiclyavailable gene expression data in MIAME format for over 66 publicly availableexperiments and also integrated tools for expression profile analysis. Finally, the GeneExpression Omnibus (GEO) (69) database at the NCBI contains data from over 300large-scale publicly available microarray and SAGE experiments, for which all data islinked into the NCBI protein, nucleotide and genomic databases.

Using these resources, it is hence possible select a number of datasets for anorganism of interest, and extract gene expression profiles for some or all genes. Proteinswhose genes exhibit very similar patterns of expression across multiple states orexperiments (70) may then be considered candidates for functional association andpossibly direct physical interaction (4-6). Gene expression analysis becomes much morereliable with more expression data. For example, genes that have high correlation across10 experiments are much more likely to be related functionally than genes correlatingacross two experiments. Gene expression data is relatively susceptible to noise, and greatcare must be taken to minimize and filter this from any analysis. This data can, however,be very powerful when combined with analyses involving regulatory networkreconstruction, and with other methods of detection of functional association andinteraction of proteins (8).

The Bayesian networks approach (see Introduction 1.3.), which combines datafrom multiple biological datasets is a useful way to minimize this noise and performreliable protein-protein interaction prediction in S. cerevisiae (8). Validation of themethod indicates that it can successfully recover large numbers of previously known


protein-protein interactions (Fig. 4) and many novel interaction predictions. The resultsof this analysis are available from the GeneCensus web site at Yale University (Table1). These predictions are remarkable as they illustrate that combining multipleindependent and noisy datasets in an intelligent way does not necessarily increase noisein the combined protein interaction predictions (assuming orthogonal error betweendatasets). This is also an excellent example of a combined computational andexperimental approach, as interactions predicted using this approach appear to be morereliable than many pure experimental approaches (8).

2.6. Tools for Protein-Protein Interaction Visualization

Network and pathway visualization tools are computer programs that can automaticallygenerate a diagram of a network or pathway. Perhaps the simplest such representation ofa protein-protein interaction network is a graph composed of nodes (proteins) connectedby edges (interactions). Some of the first visualization tools were developed for browsingmetabolic pathways. For example, a pathway drawing tool is present in the ACeDBdatabase (71) and in EcoCyc (34). In many cases these representations are clickable sothat one can select a member of a pathway or a small molecule and get furtherinformation about that entity. Many of these initial visualization tools are static, andgenerated semi-automatically. The Kyoto Encyclopedia of Genes and Genomes (KEGG)(72), BioCarta and SigPath (73) websites (Table 1), are examples of this type ofvisualization. Other more advanced methods can dynamically generate pathway diagramsfrom raw information in a biological database, such as the EcoCyc and WIT databases(see Table 1).

Recently, a number of purely automatic and general algorithms have beendeveloped for visualizing biological networks. These tools rely on a layout algorithm toorganize a graph of nodes and edges into an aesthetically pleasing layout. In graph termsthis usually means minimizing the number of edges that cross each other, and groupinggroups of nodes that are highly connected to each other. Typically, a well-organizedgraph layout will allow the user to identify global features of their data that may not havebeen previously apparent. An example layout algorithm is the Spring Embedderalgorithm. This method models the graph as a physical system where nodes are spheresconnected by springs (edges). Nodes are initially organized in a random state, and forcesbetween connected spheres (due to springs), push the system into a lower-energy morestable state. Other methods such as the Weighted Fruchterman-Rheingold algorithm (64),represent the graph as a system of nodes which exert an attractive force (similar to aspring) between nodes connected by an edge and a distance-dependant repulsive forcebetween all nodes. Additionally the weighted algorithm allows the attractive forcesbetween node to be modulated using weights, and the energy of the entire system iscontrolled using a temperature function. Other layout algorithms, can involve arrangingnodes hierarchically, in a circular fashion or in less structured formats. It is important tochoose the best layout algorithm for the type of graph being visualized. For example, ahighly connected interaction network will not assume a meaningful graph layout when ahierarchical layout algorithm is used.


Two of the most commonly used visualization tools for biological networks areBioLayout (74) and Cytoscape (75) (Table 1). Both of these tools are written using theJAVA (see Note 6) programming language and are hence portable across a wide varietyof computer environments. Both tools also allow the interactive editing of graphs,through the movement of nodes, node labeling and the ability to change the appearanceof nodes and edges. Additionally both tools can export publication quality high-resolutiongraph images (see Note 7). BioLayout utilizes the weighted Fruchterman-Rheingoldlayout algorithm, and has a number of options for graph customization, data-overlay,export and graph analysis (Fig. 5). Cytoscape provides a number of different layoutalgorithms for producing useful visualizations and a number of plugins and importoptions for representing data such as gene expression (Fig. 6). Specifically, circular,hierarchical, organic, embedded and random layouts are available. Circular andhierarchical algorithms try to lay out a network as their names suggest. Organic andembedded are two versions of a force-directed layout algorithm. Types of plugins that arecurrently available for Cytoscape include one that allows reading PSI files (see Methods2.7.) and one called ActiveModules that finds regions of a molecular interaction networkthat are correlated across multiple gene expression experiments. Both of these methodsare suitable for small to medium sized networks (less than 1000 nodes), although it maynot be long before both layout and visualization techniques become available for theanalysis of much larger graphs.

2.7. Data resources for protein-protein interactions

Current computational and experimental methods for protein-protein interactionprediction have been generating large amounts of data. It is imperative that this data bestored in a consistent and reliable way so that it may be useful for biological research. Anumber of databases are now publicly available for making this information accessible.Two of the largest and most comprehensive interaction databases now available are theBiomolecular Interaction Network Database (B I N D ) (76) and the Database ofInteracting Proteins (DIP) (77). DIP is based at UCLA and currently contains over18,000 experimentally determined protein-protein interactions (mostly from high-throughput S. cerevisiae experiments) for over 7,000 proteins in 104 organisms.Interactions in DIP are curated both manually (by expert curators) and automatically(text-mining approaches). BIND, at the University of Toronto, not only stores and curatespair-wise protein-protein interactions, but also molecular complex information andbiological pathways. Currently, BIND contains over 21,349 protein-protein and protein-DNA interactions, 1,334 molecular complexes and 8 pathways encompassing 28genomes and over 6,000 proteins.

A number of initiatives are currently underway to ensure that these data fromdifferent interaction databases are stored in a consistent and exchangeable format. TheProteomics Standards Initiative (PSI) (78) has created a standard format for the exchangeof protein-protein interaction data, while the BioPAX format aims to capture protein-protein interactions, molecular complexes and pathway information in a single consistentontology and exchange format. Access information for DIP, BIND and a number of otherinteraction databases are detailed further below (see Table 1).


3. Notes

1. Unfortunately, the online web-server for in-silico two-hybrid predictions athttp://www.pdg.cnb.uam.es/i2h/ is not presently available, although thePlotcorr program for analysis of correlated mutations is available at:http://www.pdg.cnb.uam.es/pazos/plotcorr.html

2. Pairs of close homologs (PCHs) can be defined as follows:(a) A significant BLAST hit exists for protein a in genome X and protein a’ ingenome Y. (b) A significant BLAST hit exists for protein b in genome X andprotein b’ in genome Y. (c) The genes of proteins a and b are situated within300bp in genome X. (d) The genes of proteins a’ and b’ are situated within 300bpin genome Y.

3. While BLAST is useful for many analyses of this type, it is also possible to usemore sensitive algorithms such as PSI-BLAST, HMMER or Smith-Waterman.Although these methods tend to be far more computationally intensive, the mayproduce more accurate predictions.

4. Phylogenetic profiles do not necessarily have to be binary representations. Itwould also be possible to generate profiles that express a score or expectationvalue that a homolog is present in a given genome instead of simply ‘1’ and ‘0’.

5. This type of analysis is very easy to perform using common mathematicalanalysis tools or the PEARSON function from Microsoft Excel™.

6. The Java™ environment is commonly preinstalled on many computer systems. Ifnot already installed, it can be obtained at: http://java.sun.com/

7. Graph images of protein-protein interaction networks can be exported in a numberof ways. Capturing screenshots of either application will most likely result in apoor quality low-resolution image. For publication quality images, it is generallybest to export images as a vector graphics format such as PDF.

8. The PyMOL molecular graphics package is freely available for a variety ofplatforms at: http://pymol.sourceforge.net/

Acknowledgements

The authors would like to thank Ronald Jansen for providing information about Bayesiannetwork based prediction of protein-protein interactions and the graph used for Figure 4.


Resource Type of Resource WWW Address (URL) Ref

Structural Context Interaction PredictionProtein-ProteinInteraction Server

Structure based interactionprediction

http://www.biochem.ucl.ac.uk/bsm/PP/server/ (18)

InterPreTS Structure based interactionprediction

http://www.russell.embl.de/interprets/ (33)

Genomic Context Interaction PredictionAllFUSE Gene fusions http://www.ebi.ac.uk/research/cgg/allfuse/ (51)STRING Gene Co-Localization, gene-

fusion, phylogenetic profileshttp://www.bork.embl-heidelberg.de/STRING/ (55)

WIT Orthology/phylogenetic profiles /gene co-localization

http://wit.mcs.anl.gov/WIT2/ (56)

Predictome Gene Co-Localization, gene-fusion, phylogenetic profiles

http://predictome.bu.edu/ (58)

COGs Orthology/phylogenetic profiles http://www.ncbi.nlm.nih.gov/COG/ (59)

Biological Context Interaction PredictionGeneCensus Combined predictions (bayesian

network)http://genecensus.org/intint/ (8)

Pathway DatabasesEcoCyc Metabolic pathway analysis http://ecocyc.pangeasystems.com/ecocyc/ (34)KEGG Metabolic / regulatory pathway

analysis and reconstructionhttp://www.genome.ad.jp/kegg/ (72)

SigPath Signalling pathways http://www.sigpath.org/ (73)MIPS Pathways, complexes, cellular

locationshttp://www.mips.biochem.mpg.de/proj/yeast/pathways/index.html

(79)

Protein Interaction DatabasesBIND Interactions, complexes, pathways http://www.bind.ca/ (76)DIP Database of protein interactions http://dip.doe-mbi.ucla.edu/ (77)INTACT Database of protein interactions http://www.ebi.ac.uk/intact/index.html (80)MINT Database of protein interactions http://160.80.34.4/mint/ (81)

Gene-Expression DatabasesSMD Gene expression data http://genome-www5.stanford.edu/ (67)Array Express Gene expression data http://www.ebi.ac.uk/arrayexpress/ (68)GEO Gene expression data http://www.ncbi.nlm.nih.gov/geo/ (69)

Visualization Tools for Protein InteractionsBioLayout Interaction Network Visualization http://www.biolayout.org/ (74)Cytoscape Interaction Network Visualization http://www.cytoscape.org/ (75)Table 1: Methods and databases for computational prediction of protein-proteininteractions.


Figure Legends

Figure 1: Three-dimensional structure of the T7 bacteriophage RNA polymerasecomplexed with T7 lysozyme. The multi-colored structure on the left is RNA polymerase, shownwith a transparent blue molecular surface. The lysozyme is shown on the right in grey with itsassociated transparent surface. The interaction interface is highlighted in yellow on both surfaces.This figure was produced using PDB structure 1ARO and PyMol (see Note 8).

Figure 2: Overview of genome context approaches. A) Gene neighborhood plots for eightcomplete genomes, showing a pair of genes (red and blue) which are in close physical proximityin all eight genomes. A gene fusion event between two genes (yellow and light blue) in twogenomes is also shown. B) Example phylogenetic profiles of selected genes from the previouspanel. These three pairs of genes have the same patterns of co-occurrence in all eight genomes,and may physically interact based on this evidence. C) Two protein family alignments are shownwith conserved regions highlighted (in red and blue). Correlated mutations (shown in green) arepresent in two identical sub-trees for each family, which indicates that these sites may beinvolved in mediating interactions between proteins from each family.

Figure 3: Screenshots from the STRING web resource. The left panel illustrates the STRINGrepresentation of gene neighborhood and gene fusion. The right panel shows a typicalphylogenetic profile for multiple genes and genomes. Finally, the inset shows a predicted proteininteraction map generated from gene neighborhood, gene fusion and phylogenetic profilemethods combined.

Figure 4: Bayesian network predictions of protein-protein interactions. Experimentallyvalidated gold-standard protein-protein interactions (blue and green lines) between S. cerevisiaeproteins (green dots) are shown as an interaction network. Bayesian network analysis predictionof protein-protein interaction successfully recovers a significant subset of these interactions(green lines). The gold-standard interactions are derived from MIPS and well-known complexesare annotated.

Figure 5: Example graph from BioLayout. This graph illustrates a genetic regulatory network ofE. coli genes. Genes are represented by circles (nodes) connected by regulatory interactionsrepresented by lines (edges). Nodes are colored according to biochemical pathway assignments.Nodes in the center of the graph are not labeled for clarity.

Figure 6: Example graph from Cytoscape. A number of the important features of CytoScapeare represented in this graph layout. Nodes in this case represent genes and edges representeither genetic (green, cyan) interactions, protein-protein interactions (blue) or protein-DNAinteractions (red). Nodes are colored according to the gene expression of that gene in a Gal4knockout experiment, with blue representing highly significant fold-change of a gene, and redindicating no significant fold-change. Node shapes are determined by the annotation of eachgene, diamonds for signal transduction genes, triangles for meiosis, Pol III transcription, matingresponse and DNA repair. Circles represent genes that were not assigned to any of thesecategories.


References

1. Mendelsohn, A. R., and Brent, R. (1999) Protein interaction methods - toward anendgame. Science 284, 1948-1950.

2. Eisenberg, D., Marcotte, E. M., Xenarios, I., and Yeates, T. O. (2000) Proteinfunction in the post-genomic era. Nature 405, 823-826.

3. Huynen, M., Snel, B., Lathe, W., and Bork, P. (2000) Exploitation of gene context.Curr Opin Struct Biol 10, 366-370.

4. Grigoriev, A. (2001) A relationship between gene expression and protein interactionson the proteome scale: analysis of the bacteriophage T7 and the yeast Saccharomycescerevisiae. Nucleic Acids Res 29, 3513-3519.

5. Ge, H., Liu, Z., Church, G. M., and Vidal, M. (2001) Correlation betweentranscriptome and interactome mapping data from Saccharomyces cerevisiae. NatGenet 29, 482-486.

6. Jansen, R., Greenbaum, D., and Gerstein, M. (2002) Relating whole-genomeexpression data with protein-protein interactions. Genome Res 12, 37-46.

7. Marcotte, E. M., Pellegrini, M., Thompson, M. J., Yeates, T. O., and Eisenberg, D.(1999) A combined algorithm for genome-wide prediction of protein function. Nature402, 83-86.

8. Jansen, R., Yu, H., Greenbaum, D., Kluger, Y., Krogan, N. J., Chung, S., Emili, A.,Snyder, M., Greenblatt, J. F., and Gerstein, M. (2003) A Bayesian networks approachfor predicting protein-protein interactions from genomic data. Science 302, 449-453.

9. Sussman, J. L., Lin, D., Jiang, J., Manning, N. O., Prilusky, J., Ritter, O., and Abola,E. E. (1998) Protein Data Bank (PDB): database of three-dimensional structuralinformation of biological macromolecules. Acta Crystallogr D Biol Crystallogr 54,1078-1084.

10. Chothia, C., and Janin, J. (1975) Principles of protein-protein recognition. Nature256, 705-708.

11. Gallet, X., Charloteaux, B., Thomas, A., and Brasseur, R. (2000) A fast method topredict protein interaction sites from sequences. Journal of Molecular Biology 302,917-926.

12. Korn, A. P., and Burnett, R. M. (1991) Distribution and Complementarity ofHydropathy in Multisubunit Proteins. Proteins-Structure Function and Genetics 9,37-55.

13. Young, L., Jernigan, R. L., and Covell, D. G. (1994) A Role for SurfaceHydrophobicity in Protein-Protein Recognition. Protein Science 3, 717-729.

14. Mueller, T. D., and Feigon, J. (2002) Solution structures of UBA domains reveal aconserved hydrophobic surface for protein-protein interactions. Journal of MolecularBiology 319, 1243-1255.

15. Lijnzaad, P., and Argos, P. (1997) Hydrophobic patches on protein subunit interfaces:Characteristics and prediction. Proteins-Structure Function and Genetics 28, 333-343.

16. Janin, J., Miller, S., and Chothia, C. (1988) Surface, Subunit Interfaces and Interior ofOligomeric Proteins. Journal of Molecular Biology 204, 155-164.

17. Argos, P. (1988) An Investigation of Protein Subunit and Domain Interfaces. ProteinEngineering 2, 101-113.


18. Jones, S., and Thornton, J. M. (1996) Principles of protein-protein interactions.Proceedings of the National Academy of Sciences of the United States of America 93,13-20.

19. Ofran, Y., and Rost, B. (2003) Analysing six types of protein-protein interfaces.Journal of Molecular Biology 325, 377-387.

20. Jones, S., and Thornton, J. M. (1997) Analysis of protein-protein interaction sitesusing surface patches. Journal of Molecular Biology 272, 121-132.

21. Jones, S., and Thornton, J. M. (1997) Prediction of protein-protein interaction sitesusing patch analysis. Journal of Molecular Biology 272, 133-143.

22. Lawrence, M. C., and Colman, P. M. (1993) Shape Complementarity at Protein-Protein Interfaces. Journal of Molecular Biology 234, 946-950.

23. Gabb, H. A., Jackson, R. M., and Sternberg, M. J. E. (1997) Modelling proteindocking using shape complementarity, electrostatics and biochemical information.Journal of Molecular Biology 272, 106-120.

24. Shoichet, B. K., and Kuntz, I. D. (1991) Protein Docking and Complementarity.Journal of Molecular Biology 221, 327-346.

25. Aloy, P., and Russell, R. B. (2002) Potential artefacts in protein-interaction networks.Febs Letters 530, 253-254.

26. Casari, G., Sander, C., and Valencia, A. (1995) A method to predict functionalresidues in proteins. Nat Struct Biol 2, 171-178.

27. Lichtarge, O., Bourne, H. R., and Cohen, F. E. (1996) An evolutionary trace methoddefines binding surfaces common to protein families. J Mol Biol 257, 342-358.

28. Pazos, F., Helmer-Citterich, M., Ausiello, G., and Valencia, A. (1997) Correlatedmutations contain information about protein-protein interaction. J Mol Biol 271, 511-523.

29. Zhou, H. X., and Shan, Y. B. (2001) Prediction of protein interaction sites fromsequence profile and residue neighbor list. Proteins-Structure Function and Genetics44, 336-343.

30. Fariselli, P., Pazos, F., Valencia, A., and Casadio, R. (2002) Prediction of protein-protein interaction sites in heterocomplexes with neural networks. European Journalof Biochemistry 269, 1356-1361.

31. Ofran, Y., and Rost, B. (2003) Predicted protein-protein interaction sites from localsequence information. Febs Letters 544, 236-239.

32. Aloy, P., and Russell, R. B. (2002) Interrogating protein interaction networks throughstructural biology. Proceedings of the National Academy of Sciences of the UnitedStates of America 99, 5896-5901.

33. Aloy, P., and Russell, R. B. (2003) InterPreTS: protein Interaction Prediction throughTertiary Structure. Bioinformatics 19, 161-162.

34. Karp, P. D., Riley, M., Saier, M., Paulsen, I. T., Paley, S. M., and Pellegrini-Toole, A.(2000) The EcoCyc and MetaCyc databases. Nucleic Acids Res 28, 56-59.

35. Tamames, J., Casari, G., Ouzounis, C., and Valencia, A. (1997) Conserved clusters offunctionally related genes in two bacterial genomes. J Mol Evol 44, 66-73.

36. Dandekar, T., Snel, B., Huynen, M., and Bork, P. (1998) Conservation of gene order:a fingerprint of proteins that physically interact. Trends Biochem Sci 23, 324-328.


37. Overbeek, R., Fonstein, M., D'Souza, M., Pusch, G. D., and Maltsev, N. (1999) Theuse of gene clusters to infer functional coupling. Proc Natl Acad Sci U S A 96, 2896-2901.

38. Zorio, D. A., Cheng, N. N., Blumenthal, T., and Spieth, J. (1994) Operons as acommon form of chromosomal organization in C. elegans. Nature 372, 270-272.

39. Blumenthal, T. (1998) Gene clusters and polycistronic transcription in eukaryotes.Bioessays 20, 480-487.

40. Snel, B., Bork, P., and Huynen, M. A. (2002) Genomes in flux: the evolution ofarchaeal and proteobacterial gene content. Genome Res 12, 17-25.

41. Kunin, V., Cases, I., Enright, A. J., de Lorenzo, V., and Ouzounis, C. A. (2003)Myriads of protein families, and still counting. Genome Biol 4, 401.

42. Ouzounis, C. (1999) Orthology: another terminology muddle. Trends Genet 15, 445.43. Pellegrini, M., Marcotte, E. M., Thompson, M. J., Eisenberg, D., and Yeates, T. O.

(1999) Assigning protein functions by comparative genome analysis: proteinphylogenetic profiles. Proc Natl Acad Sci U S A 96, 4285-4288.

44. Ouzounis, C., and Kyrpides, N. (1996) The emergence of major cellular processes inevolution. FEBS Lett 390, 119-123.

45. Rivera, M. C., Jain, R., Moore, J. E., and Lake, J. A. (1998) Genomic evidence fortwo functionally distinct gene classes. Proc Natl Acad Sci U S A 95, 6239-6244.

46. Marcotte, E. M., Xenarios, I., van Der Bliek, A. M., and Eisenberg, D. (2000)Localizing proteins in the cell from their phylogenetic profiles. Proc Natl Acad Sci US A 97, 12115-12120.

47. Galperin, M. Y., and Koonin, E. V. (2000) Who's your neighbor? New computationalapproaches for functional genomics. Nat Biotechnol 18, 609-613.

48. Enright, A. J., Iliopoulos, I., Kyrpides, N. C., and Ouzounis, C. A. (1999) Proteininteraction maps for complete genomes based on gene fusion events. Nature 402, 86-90.

49. Marcotte, E. M., Pellegrini, M., Ng, H. L., Rice, D. W., Yeates, T. O., and Eisenberg,D. (1999) Detecting protein function and protein-protein interactions from genomesequences. Science 285, 751-753.

50. Enright, A. J., Van Dongen, S., and Ouzounis, C. A. (2002) An efficient algorithm forlarge-scale detection of protein families. Nucleic Acids Res 30, 1575-1584.

51. Enright, A. J., and Ouzounis, C. A. (2001) Functional associations of proteins inentire genomes by means of exhaustive detection of gene fusions. Genome Biol 2,RESEARCH0034.

52. Pazos, F., and Valencia, A. (2002) In silico two-hybrid system for the selection ofphysically interacting protein pairs. Proteins 47, 219-227.

53. Gobel, U., Sander, C., Schneider, R., and Valencia, A. (1994) Correlated mutationsand residue contacts in proteins. Proteins 18, 309-317.

54. Snel, B., Bork, P., and Huynen, M. A. (2002) The identification of functionalmodules from the genomic association of genes. Proc Natl Acad Sci U S A 99, 5890-5895.

55. Snel, B., Lehmann, G., Bork, P., and Huynen, M. A. (2000) STRING: a web-server toretrieve and display the repeatedly occurring neighbourhood of a gene. Nucleic AcidsRes 28, 3442-3444.


56. Overbeek, R., Larsen, N., Pusch, G. D., D'Souza, M., Selkov, E., Jr., Kyrpides, N.,Fonstein, M., Maltsev, N., and Selkov, E. (2000) WIT: integrated system for high-throughput genome sequence analysis and metabolic reconstruction. Nucleic AcidsRes 28, 123-125.

57. von Mering, C., Huynen, M., Jaeggi, D., Schmidt, S., Bork, P., and Snel, B. (2003)STRING: a database of predicted functional associations between proteins. NucleicAcids Res 31, 258-261.

58. Mellor, J. C., Yanai, I., Clodfelter, K. H., Mintseris, J., and DeLisi, C. (2002)Predictome: a database of putative functional links between proteins. Nucleic AcidsRes 30, 306-309.

59. Tatusov, R. L., Koonin, E. V., and Lipman, D. J. (1997) A genomic perspective onprotein families. Science 278, 631-637.

60. Tatusov, R. L., Fedorova, N. D., Jackson, J. D., Jacobs, A. R., Kiryutin, B., Koonin,E. V., Krylov, D. M., Mazumder, R., Mekhedov, S. L., Nikolskaya, A. N., Rao, B. S.,Smirnov, S., Sverdlov, A. V., Vasudevan, S., Wolf, Y. I., Yin, J. J., and Natale, D. A.(2003) The COG database: an updated version includes eukaryotes. BMCBioinformatics 4, 41.

61. Altschul, S. F., Madden, T. L., Schaffer, A. A., Zhang, J., Zhang, Z., Miller, W., andLipman, D. J. (1997) Gapped BLAST and PSI-BLAST: a new generation of proteindatabase search programs. Nucleic Acids Res 25, 3389-3402.

62. Iliopoulos, I., Enright, A. J., Poullet, P., and Ouzounis, C. (2003) Mapping functionalassociations in the entire genome of Drosophila melanogaster. Comparative andFunctional Genomics 4, 337-341.

63. Smith, T. F., and Waterman, M. S. (1981) Identification of common molecularsubsequences. J Mol Biol 147, 195-197.

64. Enright, A. J. (2002) Computational Analysis of Protein Function in CompleteGenomes. Ph.D. Thesis, University of Cambridge pp. 241.

65. Enright, A. J., Kunin, V., and Ouzounis, C. A. (2003) Protein families and TRIBES ingenome sequence space. Nucleic Acids Res 31, 4632-4638.

66. Brazma, A., Hingamp, P., Quackenbush, J., Sherlock, G., Spellman, P., Stoeckert, C.,Aach, J., Ansorge, W., Ball, C. A., Causton, H. C., Gaasterland, T., Glenisson, P.,Holstege, F. C., Kim, I. F., Markowitz, V., Matese, J. C., Parkinson, H., Robinson,A., Sarkans, U., Schulze-Kremer, S., Stewart, J., Taylor, R., Vilo, J., and Vingron, M.(2001) Minimum information about a microarray experiment (MIAME)-towardstandards for microarray data. Nat Genet 29, 365-371.

67. Gollub, J., Ball, C. A., Binkley, G., Demeter, J., Finkelstein, D. B., Hebert, J. M.,Hernandez-Boussard, T., Jin, H., Kaloper, M., Matese, J. C., Schroeder, M., Brown,P. O., Botstein, D., and Sherlock, G. (2003) The Stanford Microarray Database: dataaccess and quality assessment tools. Nucleic Acids Res 31, 94-96.

68. Brazma, A., Parkinson, H., Sarkans, U., Shojatalab, M., Vilo, J., Abeygunawardena,N., Holloway, E., Kapushesky, M., Kemmeren, P., Lara, G. G., Oezcimen, A., Rocca-Serra, P., and Sansone, S. A. (2003) ArrayExpress - a public repository formicroarray gene expression data at the EBI. Nucleic Acids Res 31, 68-71.

69. Edgar, R., Domrachev, M., and Lash, A. E. (2002) Gene Expression Omnibus: NCBIgene expression and hybridization array data repository. Nucleic Acids Res 30, 207-210.


70. Eisen, M. B., Spellman, P. T., Brown, P. O., and Botstein, D. (1998) Cluster analysisand display of genome-wide expression patterns. Proc Natl Acad Sci U S A 95,14863-14868.

71. Walsh, S., Anderson, M., and Cartinhour, S. W. (1998) ACEDB: a database forgenome information. Methods Biochem Anal 39, 299-318.

72. Kanehisa, M., and Goto, S. (2000) KEGG: kyoto encyclopedia of genes and genomes.Nucleic Acids Res 28, 27-30.

73. Campagne, F., Neves, S., Chang, C., Skrabanek, L., Ram, P. T., Iyengar, R., andWeinstein, H. (2003), In Press.

74. Enright, A. J., and Ouzounis, C. A. (2001) BioLayout - an automatic graph layoutalgorithm for similarity visualization. Bioinformatics 17, 853-854.

75. Shannon, P., Markiel, A., Ozier, O., Baliga, N. S., Wang, J. T., Ramage, D., Amin,N., Schwikowski, B., and Ideker, T. (2003) Cytoscape: A software environment forintegrated models of biomolecular interaction networks. Genome Res, In Press.

76. Bader, G. D., Betel, D., and Hogue, C. W. (2003) BIND: the BiomolecularInteraction Network Database. Nucleic Acids Res 31, 248-250.

77. Xenarios, I., Rice, D. W., Salwinski, L., Baron, M. K., Marcotte, E. M., andEisenberg, D. (2000) DIP: the database of interacting proteins. Nucleic Acids Res 28,289-291.

78. Orchard, S., Hermjakob, H., and Apweiler, R. (2003) The proteomics standardsinitiative. Proteomics 3, 1374-1376.

79. Mewes, H. W., Frishman, D., Guldener, U., Mannhaupt, G., Mayer, K., Mokrejs, M.,Morgenstern, B., Munsterkotter, M., Rudd, S., and Weil, B. (2002) MIPS: a databasefor genomes and protein sequences. Nucleic Acids Res 30, 31-34.

80. Hermjakob, H., Montecchi-Palazzi, L., Lewington, C., Mudali, S., Kerrien, S.,Orchard, S., Vingron, M., Roechert, B., Roepstorff, P., Valencia, A., Margalit, H.,Armstrong, J., Bairoch, A., Cesareni, G., Sherman, D., and R., A. (2004) IntAct - anopen source molecular interaction database. Nucleic Acids Res, In Press.

81. Zanzoni, A., Montecchi-Palazzi, L., Quondam, M., Ausiello, G., Helmer-Citterich,M., and Cesareni, G. (2002) MINT: a Molecular INTeraction database. FEBS Lett513, 135-140.

computational prediction of protein-protein...

Documents